CharyZeng commited on
Commit
cd9cff2
·
verified ·
1 Parent(s): 90d8988

Update README

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -53,7 +53,7 @@ DaisyRose-27B keeps the base model's architecture, tokenizer and chat template,
53
  <sub>⁶ Averaged over 4 repetitions per prompt (300 prompts), official allenai/IFBench scorer, prompt-level loose accuracy on the final response (thinking excluded).</sub>
54
  <sub>⁷ Averaged over 8 repetitions per question (198 questions), simple-evals multiple-choice template and answer extraction, answer order reshuffled each repetition.</sub>
55
 
56
- <sub>All DaisyRose-27B runs use the base model's recommended thinking-mode settings: `temperature 1.0`, `top_p 0.95`, `top_k 20`, `min_p 0`, reasoning effort `xhigh` (the template default), no `max_tokens` cap within the 256K context. DaisyRose-27B scores are the higher of our BF16 and FP8 runs, which differ by less than one point on every benchmark where we ran both. Other models' numbers are as reported by their developers: see the sources below.</sub>
57
 
58
  </details>
59
 
 
53
  <sub>⁶ Averaged over 4 repetitions per prompt (300 prompts), official allenai/IFBench scorer, prompt-level loose accuracy on the final response (thinking excluded).</sub>
54
  <sub>⁷ Averaged over 8 repetitions per question (198 questions), simple-evals multiple-choice template and answer extraction, answer order reshuffled each repetition.</sub>
55
 
56
+ <sub>All DaisyRose-27B runs use the base model's recommended thinking-mode settings: temperature 1.0, top_p 0.95, top_k 20, min_p 0, reasoning effort xhigh (the template default), no max_tokens cap within the 256K context. DaisyRose-27B scores are the higher of our BF16 and FP8 runs, which differ by less than one point on every benchmark where we ran both. Other models' numbers are as reported by their developers: see the sources below.</sub>
57
 
58
  </details>
59